A method for improving battery utilization efficiency based on a big data battery scheduling algorithm

Abstract

The application discloses a kind of based on big data's battery scheduling algorithm promotion battery utilization efficiency's method, in turn including data reporting and collection, data collection and storage, data calculation, data back, work order distribution and battery scheduling execution five steps;Through the establishment of gray model, and then using trained gray model, predict future rental, group and battery cabinet use number of people, and according to the optimized gray model guide battery allocation and scheduling, generate work order corresponding scheduling data, work order sends distance needs battery scheduling battery cabinet nearest position of first-line dispatch personnel, first-line dispatch personnel goes to the battery cabinet position specified in work order, according to the allocation of battery in battery cabinet shown by work order information and combines scheduling path and executes battery scheduling job.The application carries out real-time monitoring and calculation to battery replacement order and battery quantity in battery replacement cabinet, and optimizes scheduling to meet the needs of rider battery replacement, while improving the utilization efficiency of battery.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of battery replacement scheduling of two-wheeled vehicles, and particularly relates to a method for improving battery utilization efficiency based on a battery scheduling algorithm. BACKGROUND

[0002] Nowadays, take-out has become a normal way of life for people, and among the delivery rider population, two-wheeled vehicle battery replacement has become a popular energy supplement method. However, in the two-wheeled vehicle battery replacement enterprise, the number of batteries in the electric cabinet is not enough, thereby affecting the battery replacement experience of the rider, or there are too many batteries in the electric cabinet, resulting in low battery utilization efficiency, thereby causing the battery replacement enterprise to be unable to make a profit, which is a major and common problem currently faced by two-wheeled vehicle battery replacement enterprises. SUMMARY

[0003] The present application provides a method for improving battery utilization efficiency based on a battery scheduling algorithm, which overcomes the deficiencies in the prior art by monitoring and calculating the number of batteries and battery replacement orders in the battery replacement cabinet in real time, and optimizing scheduling to meet the needs of riders for battery replacement while improving the utilization efficiency of batteries.

[0004] Technical solution: To achieve the above purpose, the method for improving battery utilization efficiency based on a battery scheduling algorithm of the present application comprises the following steps:

[0005] Step S1: data reporting and collection

[0006] The number of batteries and battery replacement orders in the battery replacement cabinet are detected and calculated in real time by a device with Internet of Things function, and the data is transmitted to a big data platform;

[0007] Step S2: data collection and storage

[0008] Data collection and storage are completed by real-time data access or offline synchronization;

[0009] Step S3: data calculation

[0010] After cleaning the data, a grayscale model is established, the trained grayscale model is used to predict future rental, return and battery replacement cabinet usage, and the optimized grayscale model is used to guide battery allocation and scheduling to generate scheduling data corresponding to the work order;

[0011] Step S4: data return

[0012] The scheduling data is transmitted back to the work order system, the big data platform interface is read through the work order, the work order is directly created, and the instructions are automatically sent to the front-line battery scheduling personnel;

[0013] Step S5: work order distribution and battery scheduling execution

[0014] The position information of the battery swap cabinet, the position information of the first-line dispatcher, and the allocation of the batteries in the battery swap cabinet are acquired, an optimal dispatching path is determined by using a path planning algorithm, and then the work order is automatically assigned to the first-line dispatcher who is closest to the battery swap cabinet that needs to be dispatched, the first-line dispatcher goes to the position of the battery swap cabinet specified by the work order, and performs the battery dispatching operation according to the allocation of the batteries in the battery swap cabinet displayed by the work order information and in combination with the dispatching path.

[0015] Further, in step S1, the specific steps of data reporting and collection are as follows:

[0016] a) The device collects and stores data locally;

[0017] b) The device connects to the cloud platform through a 2G / 3G / 4G / NB-IoT network;

[0018] c) The device establishes a communication connection with the cloud platform;

[0019] d) The device uploads data to the cloud platform and uses the MQTT communication protocol for communication;

[0020] e) The cloud platform receives the data uploaded by the device and performs analysis and storage;

[0021] f) The platform distributes data to the big data platform center through kafka.

[0022] Further, in step S3, the steps included in data calculation are data cleaning, gray model establishment, model fusion and optimization, model evaluation and optimization, battery allocation strategy, battery dispatching optimization, dispatching execution and feedback, performance evaluation and optimization, in turn, and specifically as follows:

[0023] 1) Data cleaning

[0024] The collected raw data is cleaned and preprocessed, noise is removed, outliers are processed, and data format conversion and unification are performed;

[0025] 2) Gray model establishment

[0026] 2.1) Determine the type of gray model: use GM(1, 1) model to capture the overall trend of data;

[0027] 2.2) Gray modeling of historical data, parameter estimation and simulation fitting;

[0028] 2.3) Model training and prediction:

[0029] The original sequence of the number of people entering and leaving the site is X^(0) = [x^(0)(1), x^(0)(2),..., x^(0)(n)], where n represents the number of historical days, X^(0) represents the sequence of the number of people entering and leaving in the past n time periods, and x^(0)(n) represents the specific number of people entering and leaving in the nth time period;

[0030] The cumulative generation sequence is X^(1) = [x^(1)(1), x^(1)(2),..., x^(1)(n)], where x^(1)(k) = ∑x^(0)(i), i from 1 to k

[0031] The cumulative reduction generation sequence is X^(2) = [x^(2)(1), x^(2)(2),..., x^(2)(n)], where x^(2)(k) = (x^(1)(k) + x^(1)(k+1)) / 2, k from 1 to n-1;

[0032] The gray action sequence is Z^(1) = [z^(1)(2), z^(1)(3),..., z^(1)(n)], where z^(1)(k) = x^(2)(k) - x^(2)(k-1)

[0033] - x^(2)(k+1), k from 1 to n-1;

[0034] The gray differential equation is z^(1)(k) + a*x^(1)(k) = b, where a and b are parameters of the gray differential equation, estimated by the least squares method;

[0035] The prediction model obtained from the gray differential equation is x^(0)(k+1) = (x^(1)(1) - b / a)*exp(-a*k) + b / a, where k is the prediction step number, and x^(0)(k+1) is the predicted value;

[0036] Use the historical data in n time periods as the training set to train the gray model and obtain the parameters of the model. Use the trained gray model to predict the future number of people using the rental, leaving, and battery swap cabinets;

[0037] 3) Model fusion and optimization

[0038] 3.1) Collect the time series data of the battery swap cabinet usage and calculate the mean and standard deviation;

[0039] 3.2) Calculate the range of three standard deviations based on the mean and standard deviation, with the upper limit being the mean plus three times the standard deviation and the lower limit being the mean minus three times the standard deviation;

[0040] 3.3) Check if the time series data falls within this range. If the data exceeds this range, it is considered an abnormal situation;

[0041] The prediction result of the gray model is fused with the prediction result of the normal distribution 3-sigma model;

[0042] GM(1, 1) prediction value: x^(0)(t+1) = (x^(1)(1)-b / a)*exp(-a*t)+b / a

[0043] Using the model fusion method, the prediction ability and weight of the two models are considered comprehensively, and the fused prediction result is used as the final prediction result of the rental, return and battery cabinet user number;

[0044] 4) Model evaluation and optimization

[0045] The prediction result of the fused model is compared with the true value, and the evaluation index, root mean square error and mean absolute error, are calculated to evaluate the prediction accuracy and accuracy of the model;

[0046] True value sequence: Y = [y(1), y(2),..., y(T)];

[0047] The prediction value sequence of the fused model: Y_pred = [y_pred(1), y_pred(2),..., y_pred(T)];

[0048] Root mean square error: RMSE = sqrt(mean((Y-Y_pred)^2));

[0049] Mean absolute error: MAE = mean(abs(Y-Y_pred));

[0050] Where: mean() represents the average value, sqrt() represents the square root operation, and abs() represents the absolute value operation;

[0051] According to the evaluation result, the model is optimized and adjusted;

[0052] 5) Battery allocation strategy

[0053] According to the real-time state and demand analysis of the batteries in the battery cabinet, the battery allocation strategy is formulated, which is as follows:

[0054] 5.1) Mark the fully charged batteries as available batteries, and mark the batteries that need to be charged as charging state;

[0055] 5.2) According to the demand and the number of available batteries, select a part of the charging batteries to meet the demand of the rental order;

[0056] 5.3) When the number of available batteries is insufficient, determine the number of recyclable batteries according to the predicted return order, and mark them as available batteries;

[0057] 6) Battery scheduling optimization

[0058] The battery is scheduled by an optimization algorithm to improve the utilization efficiency of the battery. Specifically, the battery is optimized and scheduled using a simulated annealing algorithm to balance the distribution of the battery, balance the charging and discharging times of the battery, and prolong the service life of the battery. Specifically as follows:

[0059] 6.1) Battery distribution balance degree objective function:

[0060] Assuming that the battery state S is the remaining capacity: F_balance = abs(mean(S) - max(S));

[0061] Charging and discharging times balance degree objective function:

[0062] Assuming that the battery state S is the remaining capacity: F_cycles = std(S);

[0063] Comprehensive optimization objective function:

[0064] F_total = w1 * F_balance + w2 * F_cycles, where w1 and w2 are weight coefficients for balancing the importance of the two optimization objectives;

[0065] 6.2) Define the state space and scheme of the simulated annealing algorithm:

[0066] State definition: The state represents a battery scheduling scheme, which is represented by a binary code of length N:

[0067] I = [i(1), i(2),..., i(N)], i(j) ∈ {0, 1}

[0068] Initial temperature: Set the initial temperature to control the acceptance probability in the search process:

[0069] P(I) = F(I) / Σ(F(I))

[0070] Objective function: Define the objective function to evaluate the pros and cons of the current state;

[0071] Neighborhood search: Generate adjacent states from the current state through mutation operations, such as random bit flipping or random bit replacement operations, to generate adjacent states I';

[0072] Acceptance criteria: According to the Metropolis criterion, accept the worse state to avoid falling into a local optimal solution:

[0073] p = exp((F(I) - F(I')) / T), where T is the current temperature;

[0074] Cooling strategy: control the search process to gradually converge by reducing the temperature;

[0075] Linear cooling: T = a * T, where a is the cooling coefficient;

[0076] Repeat the above steps until the maximum number of iterations is reached;

[0077] 7) Schedule execution and feedback

[0078] According to the optimization scheduling result, the allocation and scheduling strategy of the battery is executed, specifically: the scheduling result is fed back to the battery swap cabinet management system, the instructions are automatically sent to the front-line battery scheduling personnel through the work order system for actual operation, and the allocation and scheduling of the battery are monitored and recorded, so as to analyze and improve the subsequent analysis and improvement;

[0079] 8) Performance evaluation and optimization

[0080] The performance of the optimal scheduling algorithm is evaluated, the difference between the actual scheduling result and the expected effect is compared, and according to the evaluation result, the algorithm is optimized and improved, the scheduling strategy is adjusted, and the battery utilization efficiency and the scheduling accuracy are improved.

[0081] Further, in step S4, the specific steps of data backhaul are as follows:

[0082] First step: model data sending and work order state synchronization

[0083] Logic 1: When the model evaluation optimization data is generated, it is saved to mysql after being cleaned again, and sent to the message center;

[0084] Logic 2: Listen to the specified queue of the message center, and synchronize the state of the produced work order data in real time;

[0085] Second step: creation and refresh of work order

[0086] Logic 1: Listen to the specified queue of the message center, and create and refresh work orders according to the site;

[0087] Third step: work order data feedback

[0088] Logic 1: Through timing synchronization, backup the work order data to the big data computing engine according to the date dimension, and provide basic data for model training.

[0089] Further, in step S5, the specific steps of work order dispatching and battery scheduling execution are as follows:

[0090] 1) Data preparation

[0091] Get the location information of the battery swap cabinet, the location information of the front-line scheduling personnel, and the allocation of the battery in the battery swap cabinet;

[0092] 2) Path planning

[0093] The path planning algorithm composed of the shortest path algorithm combined with the genetic algorithm determines the optimal scheduling path according to the positions of the battery swap cabinets and the positions of the first-line dispatchers;

[0094] Battery swap cabinet position: C = [(x1, y1), (x2, y2),..., (xn, yn)], representing the coordinate positions of n battery swap cabinets;

[0095] Dispatcher position: P = [(x1, y1), (x2, y2),..., (xm, ym)], representing the coordinate positions of m first-line dispatchers;

[0096] Genetic algorithm representation:

[0097] Individual representation: Chromosome is represented as an arrangement, representing the scheduling path;

[0098] Individual: I = [i(1), i(2),..., i(n)], i(j) ∈ {1, 2,..., n}

[0099] Objective function: Measure the pros and cons of the scheduling path, i.e. path length;

[0100] Objective function: F(I) = d(i(1), i(2)) + d(i(2), i(3)) +... + d(i(n-1), i(n))

[0101] Selection:

[0102] Selection probability: P(I) = F(I) / Σ(F(I))

[0103] Crossover:

[0104] Sequential crossover: Select a random subsequence, and perform sequential crossover on two parent individuals I1 and I2 to generate two offspring individuals I1' and I2';

[0105] I1' = [i(1),..., i(k), i'(k+1),..., i'(n)]

[0106] I2' = [i'(1),..., i'(k), i(k+1),..., i(n)]

[0107] Mutation:

[0108] Exchange mutation: Perform exchange operation on two positions in the offspring individual;

[0109] 3) Automatic dispatch of work orders

[0110] 3.1) Establish an interface with the work order system, and automatically distribute the path planning results to the front-line dispatchers through data transmission;

[0111] 3.2) The work order system generates a work order containing detailed information of the dispatching task according to the path planning result, and automatically sends it to the corresponding front-line dispatcher;

[0112] 4) Real-time monitoring and feedback

[0113] 4.1) The front-line dispatcher receives and views the distributed work order through the mobile terminal device;

[0114] 4.2) The work order system monitors the location information of the front-line dispatcher in real time, so as to track the progress of the work order execution and the location of the dispatcher;

[0115] 4.3) The front-line dispatcher feeds back the task completion on the mobile terminal device after completing the dispatching task.

[0116] Beneficial effects: The beneficial effects of the present application are as follows:

[0117] (1) Ensure the user's battery replacement experience and improve the battery capacity taken out;

[0118] (2) Improve the asset utilization efficiency of the battery, save cash, and increase profit;

[0119] (3) Improve the dispatching efficiency and the number of dispatching per unit time of the delivery personnel. BRIEF DESCRIPTION OF DRAWINGS

[0120] The accompanying drawings are provided for the method structure diagram of the present application; Figure 1 The accompanying drawings are provided for the structure diagram of data reporting and collection;

[0121] Figure 2 The accompanying drawings are provided for the structure diagram of real-time access mode in data collection and storage;

[0122] The accompanying drawings are provided for the structure diagram of offline synchronization mode in data collection and storage; Figure 3 The accompanying drawings are provided for the structure diagram of data backhaul.

[0123] Figure 4 The accompanying drawings are provided for the structure diagram of data backhaul.

[0124] The accompanying drawings are provided for the structure diagram of data backhaul. Figure 5 DETAILED DESCRIPTION

[0125] The present application will be further described below in conjunction with the accompanying drawings.

[0126] As shown in the accompanying drawings, a battery dispatching algorithm based on big data improves the battery utilization efficiency, which comprises the following steps: Figure 1

[0127] ​​​​Step S1: Data reporting and collection

[0128] Through the device with Internet of Things function, the number of batteries in the battery swap cabinet and the battery swap order are detected and calculated in real time, and the data is transmitted to the big data platform.

[0129] As shown in the accompanying Figure 2 , the specific steps of data reporting and collection in step S1 are as follows:

[0130] a) The device collects and stores data locally;

[0131] b) The device connects to the cloud platform through 2G / 3G / 4G / NB-IoT network;

[0132] c) The device establishes a communication connection with the cloud platform;

[0133] d) The device uploads data to the cloud platform and uses MQTT communication protocol for communication;

[0134] e) The cloud platform receives the data uploaded by the device and performs analysis and storage;

[0135] f) The platform distributes the data to the big data platform center through kafka.

[0136] Step S2: Collection and storage of data

[0137] Data collection and storage are completed by real-time data access or offline synchronization.

[0138] As shown in the accompanying Figure 3 , real-time access mode

[0139] Data format:

[0140] Attribute Explanation Remark IMEI IMEI of the device Protocol type Reported protocol type Message Content reported by the device Unprocessed message Topic Source of the device heartbeat

[0141] 1. The data center confirms the reporting protocol and message channel by developing and configuring Flink tasks;

[0142] 2. The data center accepts the corresponding kafka push data;

[0143] 3. The data center stores the original data;

[0144] 4. The data center simultaneously parses the received messages according to the predetermined protocol to complete the storage process.

[0145] As shown in the accompanying Figure 4 , offline synchronization mode

[0146] 1. Configure data source and target source: configure the data source and target source for synchronization in the configuration center, including data source type, connection information, table name, etc.

[0147] 2. Define synchronization tasks: define the data sources and target sources, data columns, synchronization start and end positions, etc. to be synchronized according to business needs;

[0148] 3. Generate data synchronization scripts: generate corresponding data synchronization scripts according to task definitions;

[0149] 4. Perform data synchronization: perform data synchronization operations according to the generated scripts to synchronize the specified source data into the target data source.

[0150] Step S3: Data calculation

[0151] After cleaning the data, a gray model is established, the trained gray model is used to predict future rental, group withdrawal and battery cabinet user number, and the optimized gray model is used to guide battery allocation and scheduling, and generate work order corresponding scheduling data.

[0152] In step S3, the data calculation includes the following steps in turn: data cleaning, gray model establishment, model fusion and optimization, model evaluation and optimization, battery allocation strategy, battery scheduling optimization, scheduling execution and feedback, performance evaluation and optimization, as follows:

[0153] 1) Data cleaning

[0154] The collected raw data is cleaned and preprocessed, noise is removed, outliers are processed, and data format conversion and unification are performed.

[0155] Handling missing values:

[0156] Detecting missing values: checking the data to find fields with missing values;

[0157] Missing value filling: according to the data reported by the battery swap cabinet, for the case of large amount of user historical behavior data, use mean, median, mode to fill numerical data, and for the case of small amount of user historical behavior data, use multiple interpolation method Miceforest method to fill.

[0158] Handling outliers:

[0159] Outlier detection: detect outliers in the data through statistical analysis methods;

[0160] Outlier processing: for data determined as outliers, if there are a large number of outliers, delete them, otherwise use replacement / interpolation methods for processing.

[0161] Data format conversion and unification:

[0162] Data Type Conversion: Convert data fields to appropriate data types based on their meaning and analysis requirements, such as converting date fields to date format, and converting string fields to numerical or categorical data.

[0163] Normalization: For data fields containing units, perform unit normalization to ensure data consistency.

[0164] Data Verification:

[0165] Logical Verification: Verify data according to logical rules, such as numerical range, consistency of associated fields, etc.

[0166] Integrity Verification: Ensure data integrity, such as checking mandatory fields, primary key uniqueness, etc.

[0167] Numerical Feature Extraction:

[0168] Standardization: Standardize numerical features to convert them to standard normal distribution with zero mean and unit variance.

[0169] Bucketing: Divide continuous features into multiple discrete intervals and convert them into categorical features.

[0170] Categorical Feature Extraction:

[0171] One-Hot Encoding: For categorical features with limited values, use One-Hot Encoding to convert them into binary feature vectors, where only one position is 1 and the rest are 0.

[0172] Ordinal Encoding: For categorical features with natural ordering, use Ordinal Encoding to convert them into continuous numerical features while preserving their order information.

[0173] Time Feature Extraction:

[0174] Timestamp Processing: For features containing time information, perform timestamp processing to extract relevant time features such as year, quarter, month, day of the week, and hour.

[0175] Time Difference Calculation: Calculate the time difference between time features, such as the duration of an order.

[0176] Geographical Information Feature Extraction:

[0177] Distance Calculation: For battery swap cabinet location information, calculate the distance between two battery swap cabinets based on latitude and longitude and use it as a feature.

[0178] Geocoding: converting geographic location information into area codes, administrative divisions, and other categorical features.

[0179] Feature extraction based on domain expertise:

[0180] Extract relevant features based on the domain expertise of the problem area, such as features related to battery health status based on factors like charging characteristics, temperature, etc.

[0181] 2) Establishment of Gray Model

[0182] 2.1) Determine the type of gray model: use GM(1,1) model to capture the overall trend of the data;

[0183] 2.2) Gray modeling of historical data, parameter estimation and simulation fitting;

[0184] 2.3) Model training and prediction:

[0185] The original data sequence of the site's people vacating: X^(0) = [x^(0)(1), x^(0)(2),..., x^(0)(n)], where: n represents the number of historical days, X^(0) represents the sequence of the number of people vacating in the past n time periods, x^(0)(n) represents the specific number of people vacating in the nth time period;

[0186] Cumulative generation sequence: X^(1) = [x^(1)(1), x^(1)(2),..., x^(1)(n)], where: x^(1)(k) = ∑x^(0)(i), i from 1 to k

[0187] Cumulative decrease generation sequence: X^(2) = [x^(2)(1), x^(2)(2),..., x^(2)(n)], where: x^(2)(k) = (x^(1)(k) + x^(1)(k+1)) / 2, k from 1 to n-1;

[0188] Gray action sequence: Z^(1) = [z^(1)(2), z^(1)(3),..., z^(1)(n)], where: z^(1)(k) = a*x^(1)(k) - b, k from 1 to n-1;

[0189] -x^(2)(k+1), k from 1 to n-1;

[0190] Gray differential equation: z^(1)(k) + a*x^(1)(k) = b, where: a and b are the parameters of the gray differential equation, estimated by the least squares method;

[0191] The prediction model obtained by solving the gray differential equation is: x^(0)(k+1) = (x^(1)(1)-b / a)*exp(-a*k)+b / a, wherein k is a prediction step number, and x^(0)(k+1) is a predicted value;

[0192] The gray model is trained using historical data in n time periods as a training set to obtain parameters of the model, and the trained gray model is used to predict future rental, return and battery cabinet user numbers;

[0193] 3) Model fusion and optimization

[0194] 3.1) Collect the battery cabinet usage time series data and calculate the mean and standard deviation;

[0195] 3.2) Calculate the range of three standard deviations according to the mean and standard deviation, the upper limit is the mean plus three times the standard deviation, and the lower limit is the mean minus three times the standard deviation;

[0196] 3.3) Check whether the time series data falls within this range, if the data exceeds this range, it is considered to be an abnormal situation;

[0197] The prediction results of the gray model and the prediction results of the normal distribution 3-sigama model are fused;

[0198] GM(1,1) prediction value: x^(0)(t+1) = (x^(1)(1)-b / a)*exp(-a*t)+b / a

[0199] The model fusion (Stacking, Ensembling) method is used to comprehensively consider the prediction ability and weight of the two models, and the fused prediction result is used as the final prediction result of the rental, return and battery cabinet user number;

[0200] 4) Model evaluation and optimization

[0201] The prediction results of the fused model are compared with the true values, and evaluation indexes, root mean square error (RMSE) and mean absolute error (MAE), are calculated to evaluate the prediction accuracy and accuracy of the model;

[0202] True value sequence: Y = [y(1), y(2),..., y(T)];

[0203] The predicted value sequence of the fused model is Y_pred = [y_pred(1), y_pred(2),..., y_pred(T)];

[0204] Root mean square error: RMSE = sqrt(mean((Y-Y_pred)^2));

[0205] Mean Absolute Error: MAE = mean(abs(Y-Y_pred));

[0206] Where: mean() represents the mean, sqrt() represents the square root operation, and abs() represents the absolute value operation.

[0207] According to the evaluation results, the model is optimized and adjusted, such as adjusting parameters, improving model structure, etc.

[0208] 5) Battery allocation strategy

[0209] According to the real-time state and demand analysis of the batteries in the battery exchange cabinet, the battery allocation strategy is formulated as follows:

[0210] 5.1) Mark the fully charged batteries as available batteries, and mark the batteries that need to be charged as charging state;

[0211] 5.2) According to the demand and the number of available batteries, select a part of the charging batteries to meet the demand of the rental order;

[0212] 5.3) When the number of available batteries is insufficient, determine the number of recyclable batteries according to the predicted rental order, and mark them as available batteries;

[0213] 6) Battery scheduling optimization

[0214] Optimize the scheduling of batteries through optimization algorithms to improve the utilization efficiency of batteries. Specifically, use the simulated annealing algorithm to optimize the scheduling of batteries, balance the allocation of batteries, balance the charging and discharging times of batteries, and prolong the life of batteries. Specifically as follows:

[0215] 6.1) Battery allocation balance degree objective function:

[0216] Assuming that the battery state S is the remaining capacity: F_balance = abs(mean(S) - max(S));

[0217] Charging and discharging times balance degree objective function:

[0218] Assuming that the battery state S is the remaining capacity: F_cycles = std(S);

[0219] Comprehensive optimization objective function:

[0220] F_total = w1*F_balance + w2*F_cycles, where w1 and w2 are weight coefficients used to balance the importance of the two optimization objectives.

[0221] 6.2) Define the state space and scheme of the simulated annealing algorithm:

[0222] State definition: State represents a battery scheduling scheme, represented by a binary code of length N:

[0223] I = [i(1), i(2),..., i(N)], i(j) ∈ {0, 1}

[0224] Initial temperature: Set the initial temperature to control the acceptance probability in the search process:

[0225] P(I) = F(I) / Σ(F(I))

[0226] Objective function: Define the objective function to evaluate the goodness of the current state, for example, consider the balance of battery allocation and the balance of charging and discharging times;

[0227] Neighborhood search: Generate adjacent states from the current state through mutation operation, mutation operation is random bit flip or random bit replacement operation, generate neighborhood state I';

[0228] Acceptance criterion: According to Metropolis criterion, accept worse state, avoid falling into local optimal solution:

[0229] p = exp((F(I) - F(I')) / T), where T is the current temperature;

[0230] Cooling strategy: Control the search process gradually converges by reducing the temperature;

[0231] Linear cooling: T = α * T, where α is the cooling coefficient;

[0232] Repeat the above steps until the maximum iteration number is reached;

[0233] In the scheduling process, consider the capacity, charging speed, discharging speed and other factors of the battery, to maximize the demand of the rental order, reduce the idle and waiting time of the battery.

[0234] 7) Scheduling execution and feedback

[0235] According to the optimized scheduling result, execute the allocation and scheduling strategy of the battery, specifically: feedback the scheduling result to the battery swap cabinet management system, automatically send the instruction to the front-line battery scheduling personnel through the work order system for actual operation, and monitor and record the allocation and scheduling of the battery, so as to analyze and improve the subsequent;

[0236] 8) Performance evaluation and optimization

[0237] Evaluate the performance of the optimal scheduling algorithm, compare the difference between the actual scheduling result and the expected effect, according to the evaluation result, optimize and improve the algorithm, adjust the scheduling strategy, improve the battery utilization efficiency and the accuracy of the scheduling.

[0238] Step S4: Data return

[0239] The scheduling data transmission returns to the work order system, reads the big data platform interface through the work order, directly creates a work order, and automatically sends instructions to the front-line battery scheduling personnel.

[0240] As shown in the accompanying Figure 5 The specific steps of data return in step S4 are as follows:

[0241] First step: model data sending and work order state synchronization

[0242] Logic 1: When the model evaluation optimization data is generated, it is saved to mysql after being cleaned again, and sent to the message center;

[0243] Logic 2: Listen to the specified queue of the message center, and synchronize the state of the produced work order data in real time;

[0244] Second step: creation and refresh of work order

[0245] Logic 1: Listen to the specified queue of the message center, and create and refresh work orders according to the site;

[0246] Third step: work order data feedback

[0247] Logic 1: Through timing synchronization, backup the work order data to the big data computing engine according to the date dimension, and provide basic data for model training.

[0248] Step S5: Work order distribution and battery scheduling execution

[0249] The position information of the battery swap cabinet, the position information of the front-line scheduling personnel, and the allocation of the batteries in the battery swap cabinet are obtained, and the optimal scheduling path is determined by using a path planning algorithm. After that, the work order is automatically distributed to the front-line scheduling personnel closest to the battery swap cabinet that needs battery scheduling. The front-line scheduling personnel goes to the battery swap cabinet position specified by the work order, and performs battery scheduling work according to the allocation of the batteries in the battery swap cabinet displayed by the work order information and combined with the scheduling path.

[0250] In step S5, the specific steps of work order distribution and battery scheduling execution are as follows:

[0251] 1) Data preparation

[0252] Get the position information of the battery swap cabinet, the position information of the front-line scheduling personnel, and the allocation of the batteries in the battery swap cabinet;

[0253] 2) Path planning

[0254] The path planning algorithm composed of the shortest path algorithm combined with the genetic algorithm determines the optimal scheduling path according to the positions of the battery swap cabinet and the front-line scheduling personnel.

[0255] Battery swap cabinet positions: C = [(x1, y1), (x2, y2),..., (xn, yn)], representing the coordinate positions of n battery swap cabinets;

[0256] Dispatcher positions: P = [(x1, y1), (x2, y2),..., (xm, ym)], representing the coordinate positions of m frontline dispatchers;

[0257] Genetic algorithm representation:

[0258] Individual representation: Chromosome is represented as an arrangement, representing the scheduling path;

[0259] Individual: I = [i(1), i(2),..., i(n)], i(j) ∈ {1, 2,..., n}

[0260] Objective function: Measure the pros and cons of the scheduling path, i.e. path length;

[0261] Objective function: F(I) = d(i(1), i(2)) + d(i(2), i(3)) +... + d(i(n-1), i(n))

[0262] Selection:

[0263] Selection probability: P(I) = F(I) / Σ(F(I))

[0264] Crossover:

[0265] Sequential crossover: Select a random subsequence, and perform sequential crossover on two parent individuals I1 and I2 to generate two offspring individuals I1' and I2';

[0266] I1' = [i(1),..., i(k), i'(k+1),..., i'(n)]

[0267] I2' = [i'(1),..., i'(k), i(k+1),..., i(n)]

[0268] Mutation:

[0269] Swap mutation: Swap two positions in the offspring individual, where the path length is represented by calculating the Euclidean distance or other distance metric between two coordinate points, such as d(i, j) = sqrt((xi-xj)^2+(yi-yj)^2);

[0270] Consider the work capacity and efficiency of frontline dispatchers, as well as the allocation of batteries in the battery swap cabinet, so that the path planning can complete the scheduling task as quickly as possible.

[0271] 3) Automatic dispatch of work orders

[0272] 3.1) Establish an interface with the work order system to automatically dispatch the path planning results to the front-line dispatchers through data transmission;

[0273] 3.2) The work order system generates a work order containing detailed information of the dispatch task based on the path planning results and automatically sends it to the corresponding front-line dispatcher. The detailed information of the dispatch task includes the location of the battery swap cabinet to be dispatched, the number of batteries to be dispatched, the destination, etc.

[0274] 4) Real-time monitoring and feedback

[0275] 4.1) The front-line dispatcher can receive and view the dispatched work order through the mobile terminal device;

[0276] 4.2) The work order system monitors the location information of the front-line dispatcher in real time to track the progress of the work order execution and the location of the dispatcher;

[0277] 4.3) After completing the dispatch task, the front-line dispatcher feeds back the task completion status on the mobile terminal device, including the task execution time, battery status, etc.

[0278] The advantages of the present application are:

[0279] (1) Ensure user's battery swap experience and improve the amount of battery taken out;

[0280] (2) Improve the asset utilization efficiency of the battery, save cash, and increase profits;

[0281] (3) Improve the dispatch efficiency and number of dispatches per unit time of the delivery personnel.

[0282] In summary, the present application monitors and calculates the battery swap orders and the number of batteries in the battery swap cabinet in real time, and optimizes the dispatch to meet the needs of the rider's battery swap, while improving the utilization efficiency of the battery.

[0283] In addition, it can also improve user experience, shorten waiting time, reduce energy waste, and make two-wheeled vehicle battery swap more intelligent and efficient.

[0284] The above is only the preferred embodiment of the present application, it should be noted that for ordinary skilled in the art, without departing from the principles of the present application, can make a number of improvements and refinements, these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for improving battery utilization efficiency based on a big data-based battery scheduling algorithm, characterized in that: The method comprises the following steps: Step S1: data reporting and collection Through the device with Internet of Things function, the number of batteries in the battery swap cabinet and the battery swap order are detected and calculated in real time, and the data are transmitted to the big data platform; Step S2: data collection and storage Data collection and storage are completed by using real-time data access or offline synchronization mode; Step S3: data calculation After the data are cleaned, a gray model is established, the trained gray model is used to predict the future rental, return and battery swap cabinet user number, and the optimized gray model is used to guide the battery allocation and scheduling, so as to generate scheduling data corresponding to the work order; Step S4: data return The scheduling data are transmitted back to the work order system, the big data platform interface is read through the work order, the work order is directly created, and the instructions are automatically sent to the front-line battery scheduling personnel; Step S5: work order distribution and battery scheduling execution The position information of the battery swap cabinet, the position information of the front-line scheduling personnel and the allocation of the batteries in the battery swap cabinet are obtained, the optimal scheduling path is determined by using a path planning algorithm, the work order is automatically distributed to the front-line scheduling personnel closest to the battery swap cabinet needing battery scheduling, the front-line scheduling personnel goes to the position of the battery swap cabinet specified by the work order, and performs the battery scheduling operation according to the allocation of the batteries in the battery swap cabinet shown by the work order and combined with the scheduling path; In step S3, the data calculation comprises the following steps in sequence: data cleaning, gray model establishment, model fusion and optimization, model evaluation and optimization, battery allocation strategy, battery scheduling optimization, scheduling execution and feedback, performance evaluation and optimization, and specifically as follows: 1) data cleaning The collected original data are cleaned and preprocessed, the noise is removed, the abnormal values are processed, and the data format is converted and unified; 2) gray model establishment 2.1) determine the type of gray model: use GM(1, 1) model to capture the overall trend of data; 2.2) establish a gray model for historical data, perform parameter estimation and simulation fitting; 2.3) model training and prediction: The original data sequence of the site user return rental is X^(0)=[x^(0)(1),x^(0)(2),...,x^(0)(n)], wherein: n represents the number of historical days, X^(0) represents the sequence of the number of user return rentals in the past n time periods, and x^(0)(n) represents the specific number of user return rentals in the nth time period; The accumulated generation sequence is X^(1)=[x^(1)(1),x^(1)(2),...,x^(1)(n)], wherein: x^(1)(k)=∑x^(0)(i), i from 1 to k; The cumulative decrease generation sequence is X^(2)=[x^(2)(1),x^(2)(2),...,x^(2)(n)], wherein: x^(2)(k)=(x^(1)(k)+x^(1)(k+1)) / 2, k from 1 to n-1; The gray action amount sequence is Z^(1)=[z^(1)(2),z^(1)(3),...,z^(1)(n)], wherein: z^(1)(k)= -x^(2)(k+1), k from 1 to n-1; Grey differential equation: z^(1)(k) + a*x^(1)(k) = b, where a and b are parameters of the grey differential equation, estimated by least squares method; The prediction model obtained by solving the grey differential equation is: x^(0)(k+1) = (x^(1)(1) - b / a)*exp(-a*k) + b / a, where k is the prediction step number, and x^(0)(k+1) is the predicted value; Use the historical data in n time periods as the training set to train the grey model, obtain the parameters of the model, and use the trained grey model to predict the future rental, return and battery cabinet usage; 3) Model fusion and optimization 3.1) Collect the battery cabinet usage time series data and calculate the mean and standard deviation; 3.2) Calculate the range of three standard deviations according to the mean and standard deviation, the upper limit is the mean plus three times the standard deviation, and the lower limit is the mean minus three times the standard deviation; 3.3) Check if the time series data falls within this range, if the data exceeds this range, it is considered an abnormal situation; Fuse the prediction results of the grey model and the prediction results of the normal distribution 3-sigma model; GM(1,1) prediction value: x^(0)(t+1) = (x^(1)(1) - b / a)*exp(-a*t) + b / a Use the model fusion method, consider the prediction ability and weight of the two models, and the fused prediction result as the final rental, return and battery cabinet usage prediction result; 4) Model evaluation and optimization Compare the prediction results of the fused model with the true value, calculate the evaluation index, root mean square error, mean absolute error, and evaluate the prediction accuracy and accuracy of the model; True value sequence: Y = [y(1), y(2),..., y(T)]; Prediction value sequence of the fused model: Y_pred = [y_pred(1), y_pred(2),..., y_pred(T)]; Root mean square error: RMSE = sqrt(mean((Y-Y_pred)^2)); Mean absolute error: MAE = mean(abs(Y-Y_pred)); Where: mean() represents the average value, sqrt() represents the square root operation, and abs() represents the absolute value operation; According to the evaluation results, optimize and adjust the model; 5) Battery allocation strategy According to the real-time state and demand analysis of the battery in the battery cabinet, formulate the battery allocation strategy, as follows: 5.1) Mark the fully charged battery as available battery, and mark the battery that needs to be charged as standby charging state; 5.2) According to the demand and the number of available batteries, select a part of the standby charging battery to charge to meet the demand of the rental order; 5.3) When the number of available batteries is insufficient, determine the number of recoverable batteries according to the predicted return order, and mark them as available batteries; 6) Battery scheduling optimization The battery is scheduled by an optimization algorithm to improve the utilization efficiency of the battery. Specifically, the simulated annealing algorithm is used to optimize the scheduling of the battery, balance the allocation of the battery, balance the charging and discharging times of the battery, and prolong the service life of the battery. The specific steps are as follows: 6.1) Battery allocation balance degree objective function: Assuming that the battery state S is the remaining capacity: F_balance = abs(mean(S) - max(S)); Charging and discharging times balance degree objective function: Assuming that the battery state S is the remaining capacity: F_cycles = std(S); Comprehensive optimization objective function: F_total = w1 * F_balance + w2 * F_cycles, where w1 and w2 are weight coefficients used to balance the importance of the two optimization objectives; 6.2) Define the state space and scheme of the simulated annealing algorithm: State definition: The state represents a battery scheduling scheme, which is represented by a binary code of length N: I = [i(1), i(2),..., i(N)], i(j) ∈ {0, 1} Initial temperature: Set the initial temperature to control the acceptance probability in the search process: P(I) = F(I) / Σ(F(I)) Objective function: Define the objective function to evaluate the pros and cons of the current state; Neighborhood search: According to the current state, generate adjacent states through mutation operations, such as random bit flipping or random bit replacement operations, to generate adjacent states I'; Acceptance criterion: According to the Metropolis criterion, accept worse states to avoid falling into local optimal solutions: p = exp((F(I) - F(I')) / T), where T is the current temperature; Cooling strategy: Control the search process to gradually converge by reducing the temperature; Linear cooling: T = α * T, where α is the cooling coefficient; Repeat the above steps until the maximum number of iterations is reached; 7) Scheduling execution and feedback According to the optimization scheduling result, the allocation and scheduling strategy of the battery is executed, specifically: the scheduling result is fed back to the battery replacement cabinet management system, the instruction is automatically sent to the front-line battery scheduling personnel through the work order system for actual operation, and the allocation and scheduling of the battery is monitored and recorded for subsequent analysis and improvement; 8) Performance evaluation and optimization The performance of the optimal scheduling algorithm is evaluated, the difference between the actual scheduling result and the expected effect is compared, and according to the evaluation result, the algorithm is optimized and improved, the scheduling strategy is adjusted, and the battery utilization efficiency and scheduling accuracy are improved.

2. The method of claim 1, wherein the method is characterized by: In step S1, the specific steps of data reporting and collection are as follows: a) The device collects and stores data locally; b) The device connects to the cloud platform through 2G / 3G / 4G / NB-IoT network; c) The device establishes a communication connection with the cloud platform; d) The device uploads data to the cloud platform using the MQTT communication protocol; e) The cloud platform receives the data uploaded by the device and performs analysis and storage; f) The platform distributes data to the big data platform center through kafka.

3. The method as claimed in claim 1, wherein the battery scheduling algorithm based on big data improves the battery utilization efficiency. In step S4, the specific steps of data back transmission are as follows: First step: Model data sending and work order state synchronization Logic 1: When the model evaluation is optimized, the data is generated after cleaning, and is saved to mysql and sent to the message center; Logic 2: Listen to the specified queue of the message center, and synchronize the production order data state in real time; Second step: creation and refresh of the order Logic 1: Listen to the specified queue of the message center, and create and refresh the order according to the site; Third step: order data feedback Logic 1: Through timing synchronization, backup the order data to the big data computing engine according to the date dimension, and provide basic data for model training.

4. The method of claim 1, wherein the method further comprises: In step S5, the specific steps of order dispatching and battery scheduling execution are as follows: 1) Data preparation Obtain the location information of the battery swap cabinet, the location information of the front-line dispatch personnel, and the allocation of the batteries in the battery swap cabinet; 2) Path planning Use the path planning algorithm composed of the shortest path algorithm combined with the genetic algorithm to determine the optimal scheduling path according to the location of the battery swap cabinet and the location of the front-line dispatch personnel; Battery swap cabinet location: C = [(x1, y1), (x2, y2), …, (xn, yn)], representing the coordinate positions of n battery swap cabinets; Dispatch personnel location: P = [(x1, y1), (x2, y2), …, (xm, ym)], representing the coordinate positions of m front-line dispatch personnel; Genetic algorithm representation: Individual representation: Chromosome is represented as an arrangement, representing the scheduling path; Individual: I = [i(1), i(2), …, i(n)], i(j) ∈ {1, 2, …, n} Objective function: Measure the degree of scheduling path, i.e. path length; Objective function: F(I) = d(i(1), i(2)) + d(i(2), i(3)) + … + d(i(n-1), i(n)) Selection: Selection probability: P(I) = F(I) / Σ(F(I)) Crossing: Sequential crossover: Select a random subsequence, and perform sequential crossover on two parent individuals I1 and I2 to generate two offspring individuals I1' and I2'; I1' = [i(1), …, i(k), i'(k+1), …, i'(n)] I2' = [i'(1), …, i'(k), i(k+1), …, i(n)] Mutation: Exchange mutation: Perform exchange operation on two positions in the offspring individual; 3) Automatic dispatch of order 3.1) Establish an interface with the order system, and automatically dispatch the path planning result to the front-line dispatch personnel through data transmission; 3.2) The order system generates an order containing detailed information of the scheduling task according to the path planning result, and automatically sends it to the corresponding front-line dispatch personnel; 4) Real-time monitoring and feedback 4.1) The front-line dispatch personnel receive and view the dispatched order through the mobile terminal device; 4.2) The order system monitors the location information of the front-line dispatch personnel in real time, so as to track the progress of order execution and the location of the dispatch personnel; 4.3) The front-line dispatch personnel feed back the task completion status on the mobile terminal device after completing the scheduling task.

Images